I think AIF files use ID3 tags to store metadata whereas Flac files use Vorbis comments. I have seen huge metadata blocks in Flac files (downloaded from HDTracks), much larger than actually required to contain the album art. One would think that removing the album art would remove the metadata block containing it but it doesn’t and player software has to read through the whole block to see if there is anything there. The metaflac command line utility can show the metadata blocks in Flac files and can be used to remove large empty album art blocks.
The workaround that James proposed should work in all cases as it does two things to alleviate the problem: removes any large Flac metadata blocks and removes the need for the player unpack the Flac data. While neither of these tasks should be a challenge for any modern processor it appears that DCS may have dug themselves into a bit of a hole with their player software design when dealing with 192/24 files.
To expand upon this a bit further, and this is speculation on my part based on James’ explanation of the problem, the Bartok manual and my own development of a software player over the last 17 years: I think the issue is that the DCS player does not look at the next track to play in advance but waits until it is due to play to read the metadata and deal with it. It uses a buffer to facilitate this but that buffer appears to be a fixed size. The manual says it is .72 seconds for 44.1k files down to .16 seconds for 192k files. With a high sample rate Flac file with an abnormally large metadata block it probably simply runs out of time to process the metadata before the buffer empties leaving silence until the flow of data resumes.
IMO a better design is to have a buffer that is larger for higher sample rates so that the buffer time remains constant and better still to do all metadata processing for the next track in advance so that there is very little to do on track transition. To retrofit either of these options to the DCS player would be non trivial since they are fundamental parts of player management and may not even be possible if there are memory constraints.
If preprocessing of the next track were implemented in the player this could also fix the DSD playback issue and facilitate implementation of gapless playback for UPnP controllers (the UPnP specification provides an optional next URL command which is not currently implemented in the DCS player).
Any such changes would require a very significant amount of testing and I suspect they will not happen soon and possibly not in the current generation of players.