If you backup your catalog regularly, you don't need the metadata box checked. Alternately If you remember to hit the [Ctrl/Cmd][S] when you are done with your file edit (to save your work done on the file) you don't need auto-write turned on either.
I had it turned on about three years ago for about 10 minutes and decided that was no way to work. Others may disagree.
To answer your question, in my world, no.
Locking for 5 minutes is abnormal. With auto-write xmp enabled it should just tag the files for xmp writing to be done leisurely in the near future (in the background) and not interfere with anything, whether it's trying to do it for every incremental movement or not.
That said, I recommend leaving auto-write xmp off anyway. Why?
You've already got current settings in the catalog. If you only save
xmp manually, after milestone editing and inspection of images,
then you are guaranteed xmp does NOT included the same goofs
your catalog does (or might). If you never goof, then please identify
home planet, cause it sure ain't Earth, and you sure ain't human..
So, my advice? - have a "finalization" procedure - when done editing
images, for now anyway:
* inspect image edits..
* save XMP.
That way, if ever you need to fall back to a "milestone" state which isn't
in backup catalog, you can just read XMP.
The primary reason I thought I'd try the auto write to xmp is because, when I do a large batch of writing metadata to file after a week or two or editing, many of them will still come up as not up to date the next time I look & I must go to the additional step of writing from file back to the catalog. It doesn't seem that Adobe is ever going to fix this bug & I thought I could avoid it by automatically writing to xmp.
Fair enough, and (theoretically) you should be able to enable that option with impunity - most Lr users can, I think (conclusion based on educated guess-work).
But of course working around bugs is just part of working .
I turn the xmp status indicators off and use other means for checking metadata status (if there is any question), since Lr's metadata status detection is flaky.
I suppose another possible bug work-around would be a 2-part sequence when manually saving metadata:
1. save metadata
2. read metadata
then status should always be correct - whether you hit that bug or not, right?
Different strokes for different folks I guess.