Uhhargh! I just checked the new biomart 0.8 web-interface, yes it looks more fancy with web 2.0 widgets and stuff but I think usability cannot keep up and the interface seems cluttered.

Maybe that's only because I need to adapt, let's see. Also, I agree that the flank functionality has changes somehow. I could swear (I could be wrong ofc) it was possible to retrieve both upstream and downstream flanks in one go. Now, you can either get upstream or downstream at once, otherwise I get:

Validation Error: For this sequence option choose upstream OR downstream flanking sequence, NOT both.

Anyway, it still works, you just have to run two queries and concatenate the fasta files. And if you don't want to use the new interface, you can use ensemble biomart which is still on the old interface (and I hope they don't migrate....). Here is a link to the query that yields upstream flanks in ensembl biomart. (the functionality to retrieve a stable URL or XML query do document what one was doing, hasn't made it into the new interface either (or is hidden very well))

I'd go into NCBI's FTP interface to download all genomes' GFF and GenBank files. You can then parse the GFF for CDS and their start/stop coordinates. For each start/stop and given a flanking size, extract the genomic sequence.

If you are looking for a GUI to do this, I am not sure if it exists and so you'd have to program it.

The BioMart web interface really is not that well-suited for these genome-wide queries, I am afraid. I would recommend using either the BioMart API or the Ensembl Perl Core API. If you use the former, you can directly use the Perl code that you can retrieve using the [Perl] button on the BioMart web interface once you have defined your query. If you use the latter, you have to write the code yourself, but it shouldn't take too much time as this is a very simple query.