It affects everything, no matter what format the source is.
In my analysis, I split all the samples in the song up into 2048 partitions.
When I get audio from the decoder, it arrives in chunks of a few thousand samples.
[_partition #0_______________________][_partition #1_______________]...
[_chunk #0_____][_chunk #1__][_chunk #2______][_chunk #3___][_chunk #4____]...
abcdefghijklmnopabcdefghijklmabcdefghijklmnopqabcdefghijklmnabcdefghijklmno...
The problem occured when a chunk was part of more than one partition, like chunk #2 is in the example above. What ended up happening was that the beginning of the chunk was replicated into the second partition, instead of the remainder.
...________][_partit...
...[_chunk #2______]...
...abcdefghijklmnopq...
became ...________][_partit...
...[_chunk #2______]...
...abcdefghiabcdefgh...
, which irrecoverably lost the jklmnopq information contained in chunk #2's second half.
The only way to not have experienced this is if the chunk size of the decoder aligned perfectly with the partition boundary, but that's a lucky corner case and near impossible.
I hope this explains the underlying problem. If you're fine with the replication of a miniscule part of the samples and a loss of the samples it clobbers, you can continue using the waveforms you have. If you want perfection, regenerate them. Don't forget that the component remembers that you've told it to scan something, so it continues when you restart the player.