Comment 82 for bug 1690719

(In reply to Jonas Ådahl from comment #62)
> There are plans to fix this. When it comes to the responsiveness of just the
> mouse cursor, we'll probably move libinput processing to its own thread (as
> is done on recent versions of Xorg) where it can directly update the
> hardware cursor without waiting for main drawing loop. This would probably
> involve moving the libinput backend from clutter to mutter, and adding
> mutexes shared by the KMS code and libinput backend. Doing this will not
> affect latency issues related to clients receiving input events; that'll
> require much larger changes, such as splitting up the shell UI into a
> separate process.

"such as splitting up the shell UI into a separate process." For me, this is the solution other things are patches... I have stumbled upon this a lot of times by doing an extension. To reduce the impact, I had to split the task into several parts that would be processed in a chain of idle subprocess. Thats give to mutter a chance to process the pending tasks.