Comment by defanor
> but what would be the benefit of having more pages fail to render?
I think those benefits are quite similar to having more programs failing to run (due to static and strong typing, other static analysis, and/or elimination of undefined behavior, for instance), or more data failing to be read (due to integrity checks and simply strict parsing): as a user, you get documents closer to valid ones (at least in the rough format), if anything at all, and additionally that discourages developers from shipping a mess. Then parsers (not just those in viewers, but anything that does processing) have a better chance to read and interpret those documents consistently, so even more things work predictably.
Sure, authoring tools should help authors avoid mistakes and produce valid content. But the browser is a tool for the consumer of content, and there is no benefit for the user if it fails to to render some existing pages.
It is like Windows jumping through hoops to support backwards compatibility even with buggy software. The interest of the customer is that the software runs.