I have experience with three motion control system real time:

1: Synqnet: It´s good but not the best, in my experience. The control is on a PC with a not real time system.

2: B&R: I like this true real time controller, and all the power that you can use from it.

3: Sigmatek: I like this true real time controller, it´s economic in comparation with other systems.

What did you use? Do you like?
If you have any question on one, pleased, ask.
If you know about other , pleased, tell us.

Best regards.

You're 100% correct, direct Windows-based control is absolutely not real-time control. There's too much happening that's non-deterministic, Windows doesn't care about what priority you feel your motion control task has. "Need to update your mission-critical PID loop? Sorry, I need to poll the print spooler, your request will processed in the order it was received."
I've used Parker controllers in the past and also controlled Kinetix drives directly from either Siemens or AB PLCs. These days all I use is National Instruments cRIO (with or without FPGA) systems and NI's LabVIEW Real-Time Module. It's more expensive than a PLC-based system, that's for sure. But, the flexibility, wide hardware selection, and ease of use usually outweigh the cost.
NI also offers the ability to import models of your motion system from Solidworks and and incorporate them seamlessly into your project using NI SoftMotion. You can then run simulations and watch your system move, tune it, and analyze the results. Pretty slick, but again, not cheap.

SynqNet is not really Windows based. It uses dedicated FPGA for Real-time Ethernet and DSP for control loop. As far as I know SynqNet is more or less abandoned by DHR in favor of EtherCAT.

Essentially, all Real-time Ethernet protocols uses similar scheme - FPGA or dedicated ASIC for on-the-fly manipulation on frames.
SERCOS III has technical advantages over few others, while Ethercat has stronger market grip.

Motion controller (I’m speaking about bus-based, not PWM or Analog commands) usually do not close servo loops (SynqNET does!) and can handle certain real-time jitter of operating system.

With SERCOS II, SERCOS III and Ethercat in position loop you can easily handle jitter of 50-100 msec at cycle time of 1-2 msecs, under condition that CPU use is not used above 60%. Please note that that jitter is in software, while network must be sub microsecond accurate.

I do not think that Windows, without real-time extension or dedicated DSP board, can handle motion control tasks.

Excellent post from Rasty! - must be sign out that we discuss bus -based motion controllers because some no bus-based as Galil for example are more Real Time really. And for applications with time critical motion control tasks for some axes such centralized motion controllers with bus-based option will be preferred - ACS or Baldor with CANOpen Master option.

EtherCAT communication between Elmo drives and Elmo Host controller (GMAS).
EtherCAT is a real time communication (few nanoseconds jitter) together with distributed control in drives you can get high performance.
Elmo gantry communication is done drive to drive every 50 microseconds

Obviously, Elmo is know for its good motion control product.
Ethercat provides sub microsecond jitter, on the other hand why having modern real-time network we still need proprietary link between drives? Answer is in limitations of Ethercat (please correct if I'm wrong!) Unlike SERCOS III, Ethercat does not offer standard slave-to-slave communication. SERCOS network can share data between slaves at 32 usec cycle time, one-to-one, many-to-one, one-to-many.
To be honest, not all drives support that, but have an option at least .

I think it is really application oriented question, such as what kind of level of real-time and sychronzation you are looking for, or pre-defined indutry-oriented feature, programming enviroment and languages preferred. Synqnet is dedicated designed for network multi-axis motion control, and just think that for 16axis servo, you can still get 8 to 16KHZ update rate (deterministics motion cycle), for one axis only, you can get 48KHZ (much more than the conventional servo drive PWM frequency). And with the faster CPU come out, the update rate can be faster. The Synqnet product itself has quite excellent tools like filter simulation design and even Mechaware that you can use this commecial controller for quite flexible and challenging in-house controldesign. But will you use this higher performance controller for packaging? Of course, not, you may like to choose maybe PLC based like IEC61131 programming and EtherCAT that you can connect many nodes, not just servo/step, but maybe sensors, inverters so on. You see, they are for different market focus.

btw, Rasty is correct. Synqnet is using FPGA for its network connection for specific design Ethenet network. and if you need real-time OS for PC, you can select Vxworks, RTX etc.

Something real intreast is recently we see some soft motion controller come out, and it says can replace current certain high performance network motion controller. Since more core and faster processor come out that this software can use one of that processor to taking care for motion control only.

It is always amazing how quickly a discussion about real-time turns into a networking discussion. EtherCAT, Sercos, SynqNet, Profinet, PowerLink all provide low-jitter cyclical data exchange (as does Firewire and USB by the way). That is required for digital real-time data sampled systems. Beyond that it is all about the actual master and slave devices you connect to it as to what kind of performance you will get. Whether you close servo loops inside the slave versus over-the-network is highly application dependent. I see many more mistakes made at the level of sizing motor torque and selecting proper feedback resolution then in the control architecture. In the end the physical time constants one deals with in motion control are in the millisecond and higher range. At some point, throwing more speed at it will not make a difference.