Hi
It seems this topic has come around again, so I will add my 2 cents worth:-)

I come from an Oracle background where PL/SQL has a much broad set of functionality than T-SQL - probably because you can write CLR for anything missing.  In the Oracle databases we developed it was mostly "stage and then process" within the database, this meant a common set of logic was used for all sources of the information (multiple file formats, directly storage from web pages, and others).  Given this philosophy when we started moving to SQL Server we followed the same development processes - which was naive, as they are different even though they are both database systems.

So we built CLRS for the following tasks:
1) calling out of the database to remote web sites (secure pages only)
2) logging to file
3) binary file generation - so the file looked like it was generated from a 1990's C application
4) communication back to the Oracle database

While these all worked, some better than others, we have been moving away from this to using tools better suited to the job e.g. custom applications/services or SSIS tasks.

So would I write a CLR again - "maybe" as that would depend on the reason/task, rather than the current philosophy.