95% of the USB KVMs on the market rely on USB's hot-attach-detach plug&pray. This results in disconnect-reconnect events being seen by each host, with a minimum of 3-5 seconds for bus re-enumeration on EVERY switch operation. If your switch is also a hub, or supports devices other than K/M ("peripheral sharing"), it is almost certainly a simple non-spoofing switch.
If you only switch occasionally, this might be ok. If you frequently switch (as between RISC workstation and PC), it's unacceptable.
4% of USB KVMs are actually just simple mechanical switches. They have all the same problems as above, plus, striclty speaking, they violate USB rules, and can have nastly malfunctions if they don't have correct precedence for power-on/off (as USB connectors do).
1% of USB switches, just now entering the market, have "spoofing", aka "sniffing", aka emulation of devices. A correctly implemented spoofing switch never lets any host see a disconnect. Designing this is far from trivial. USB is complex, and the varieties of keyboards and mice are effectively infinite if you're trying to chase them all for purposes of emulation.
In addition to the IOgear mentioned above, there is only one other spoofing switch that I'm aware of, and it isn't fully debugged yet.
In the meantime, I recommend:
- Use a quality USB(host) to PS/2(device) adaptor, such as the Y-Mouse USB (www.ymouse.com) on each USB host.
- Use a PS/2 switch.
Wait for this USB KVM situation to get sorted out. And it will need to soon, as PS/2 ports will be vanishing from new machines next year (and already have from many 64-bit platforms).