SLA's by channel type can help too...
To what extent do you apply acceptable SLA's to your planning for multichannel. Having deployed about 16 large scale omnichannel solutions across Three, T-Mobile, Orange, EE, Vodacom, Safaricom and others we were very careful to map the acceptable SLA's to the WFM/WFO planning templates. These are the rules of thumb that worked well; heavily tainted by the needs of your business strategy of course;
Twitter - 20 minutes for public or Direct Message responses
Facebook - 8 hours for wall posts
Facebook - 20 minutes for Messenger
SMS - 20 minutes
eMail - ????? Take your pick, some 72 hours; others 8 hours....!!
Communities tended to be tainted heavily by your 'moderator' audience. If you are using something like Lithium then the community will have its own frontend SLA's driven by moderator 'competition' and fallback to customer operations intervention only when it got heated, had no response in timeframe X and so on.
Factor those; then factor concurrency.
Are you routing omnichannel to eServices CSR's that have voice to callback; but are not on the primary voice queues? In some of our solution designs we queued eServices to agents on voice queues and pulled all outstanding eServices transactions and routed them alongside any immediate voice call; there's no point sending a voice call to a voice agent without them seeing that the customer has shouted on Twitter X times since 9am this morning....
I don't know which 'platform' you are on - but on ones like Genesys you should not be dumping your queues. If your SLA's are clearly defined then your queue should have a parking queue for those inside SLA and an automatic routing override for interctions that are close to SLA. You should also pull out all queued interactions when targeting a blended agent seat with any other form of eServices transaction in queue because they are related 95% of the time.
Hope some of this helps.