You're right, this is where per-mailbox spam filtering kicks in. But for that to work you need to create a programmatic mailbox (we have APIs for that) which allows us to collect your spam&ham datasets.
We do have a few customers who're using per-mailbox spam training via our "Beta API access" program, but we have not made the decision to launch it as a product yet. If this is something you may be interested in, reach me via email in my profile.
Though one company's spam is another company's ham, so I wonder if this would work well enough across all your clients?