The plugin interface should support using the queue. Since the events are queued it sort-of does currently, however I suggest using it explicitly.
This should allow plugin support to grow while maintaining near real time in the interface.
I also propose enabling the possibility to re-queue jobs should they fail (for a configurable amount of time).
Thoughts along the lines of;
Clearly some things we'd never want to re-try, other things we probably do. The network stack can be a little picky and I don't think an environment issue outside of sentry should cause it to not send out notices.
We should probably have some form of weight to ensure the top of the queue doesn't get clogged with crappy jobs.
Via a mix of general uselessness and random slowness I think it's a sensible idea for the long term.