I used to work there. The scanned returns probably make up a bit, but they simply have lots of data (e.g. last 10 years of tax returns for every taxpayer, bunch of data matching DBs, enormous maze-like website, maps to buried treasure etc. etc.)

And you're correct, their audit programmes all produce shitloads of data, all of which must be retained for legal record keeping (Archives Act).

EDIT: come to think of it, the biggest space eater would be scans of documents sent in by folks who have been audited...

Scanned information returns? They're many orders of magnitude larger than their declarative (e.g. PDF) or plain text equivalent.