Yeah, this is a bit tricky.
Normally, a single OME instance would be just fine for this number of devices. If you have the opportunity to try it out as a pilot and see how it looks, that would be good.
But, unless the connections to your various DCs are rock-solid and response times to pings are repeatedly sub-second, you may end up needed separate OMEs. Traffic depends on several variables (OME version, server generation, server options, protocol used, etc.). Maybe you should use a packet capture tool to estimate amount of data…start/end capture before/after each (discovery/inventory/status) poll.
We have a lot of whitepapers published already, but a sizing whitepaper would be a good one to add, so I'll take note.
Best regards,
Rob