I don't know how I did that. I'd written up three paragraphs describing this in detail (including confirming that WAL files were being deleted - thanks, Scott) but I must have inadvertently hit the delete key or something because there's *nothing* following the first open parenthesis in my original post! What followed that open parenthesis was the first of very many details.

I'm pressed for time at the moment, but will update later. In brief, vacuum full and reindex did nothing. A pg_dump directly to another host had one db shrink from 48Gig to approx 432MB. In the pg_data directory for that db, fully 43 Gig was consumed and bytes consumed did not change even after truncating tables and re-vacuum. Nothing untoward in any of the log files, the only database on this cluster displaying this goofiness was the aforementioned, nothing out of place, etc. I know that's not much better than no details, but it's all I have time for at the moment. I'm inclined to believe (since Java developers with little to no database experience were allowed to manage this server) that somebody did something in that data directory they thought was smart.