From the specs, there's a frame or two delay on non-referenced sources. I've never seen enough of a delay caused by the 1M/E to cause a problem. I do run analog audio from a separate mixer into the ATEM and then use it's analog outputs or embedded audio, which may (or may not) be doing the appropriate audio delay.
I have used HDMI for a cable. At 1080i, however, I can only reliably use a 25' HDMI cable. Any longer and it starts losing sync. Some cameras have a 'stronger' HDMI cable than others and can go longer.
As an alternative, I use a pair of Monoprice HDMI over Cat6 extenders (about $40), and 190' of Cat6 cable (requires two cables) to extend about 200' total from the ATEM.
At 1080i you can get decent runs out of cheap coax, the cable would probably work out cheaper than the dual run of cat6 and go straight into the SDI inputs on your switcher, freeing up some HDMIs for other tasks.
We're planning to use that and Component - SDI to bring our cameras in, its worked well for us.