Okay, in that case I'll repeat here what I said to the person who contacted us through support. The user was kind enough to record what was seeing (twice!) and upload it, making our investigation a lot more effective:
"Thanks for the recording, this makes it clearer. What you are observing actually appears to be normal (ie. working as designed).
When you are at a certain zoom level, and you zoom-in further than it, initially you get the same tile you were previously looking at, except now it is stretched and therefore blurrier. This lasts a few seconds while the closer-zoom tiles are being fetched, either from the local cache or from the remote server. Once the tiles are at hand, they replace the original tile.
Now, if you zoom-out back again, initially you are still viewing the same higher-resolution tiles, except they are now shrunk and therefore extremely crisp-looking. This is the same thing that GPUs do automatically when they apply antialiasing on a scene (ie. render the entire scene at a much higher resolution and then scale it down to the target screen resolution in order to produce a very sharp-looking image). However, this is undesirable for the map engine as it is inefficient/wasteful (at that moment we are using 4 tiles at the same screen space where one single tile, stretched normally, would suffice. So, very shortly, the 4 higher-zoom tiles are again replaced by one lower-zoom tile. The tile is of course less crisp than the previous 4 ones. This is what you are observing, and it is in fact working as designed.
Thanks again for taking the time to produce and share your recording, otherwise it would have been extremely difficult to understand what you are seeing."