A couple of weeks ago they released a graph suggesting that 78% of all users use twitter.com (http://blog.twitter.com/2010/09/evolving-ecosystem.html)
If you look carefully, they explain that this 78% is the percentage of people that have logged into twitter.com at least once in the last 30 days. Unfortunately this is not very useful since there are a multiple things that can only be done on twitter.com and not in 3rd party clients, either because the functionality is missing from the API or is broken (they started to fix some of these though). So even if people use a 3rd party client most of the time, they would still log into twitter.com every once in a while and thus will be counted as a twitter.com user.
A more useful statistics would be the percentage of users who use any given client as their main way of interacting with Twitter. Of course twitter.com would still be the biggest client, but not with 78% of all users, more likely somewhere between 40-50% (based on stats by twittersource.info, tdash.org and stats of Hungarian users I've collected).
As for the original post, we're running a web-based client and twitter.com is one of the most important referers so I'll have to agree that removing the "via client" attribution will hurt developers in a big way. (Yeah, I know, they are not removing it just making it a lot less prominent)