I'd put it this way: if a player wants to be bumped up, it's much easier to do so by playing up than by playing at level. In fact, a bump-up is almost guaranteed when playing up, nearly irrespective of the actual results.
This is due to the way the algorithm works: a player's DNTRP will almost always go up when playing someone with a substantially higher DNTRP (by "substantially higher" I mean 0.25+ range) - even if the "lower" player lost to the tune of 6-2 6-3. While that same player beating someone at-level may not raise their DNTRP; it may in fact lower it. In other words, for a player at the higher range of a given rating, it's very difficult to raise their DNTR by playing at-level, as this basically requires blow-out wins against players at the lower range of that rating, and substantial (score-wise) wins over players with similar ratings.
My results are a case-in-point. 2010: 0.750 W/L at 3.5, with lots of singles wins in 6-2 6-1 range = no bump-up. 2011 - playing up exclusively: 0.350 W/L, with most wins in competitive matches, and a few less-than competitive (score-wise) losses = bump up.
Exact same picture for another guy on my team who followed the same route: excellent results in 2010 at 3.5, no bump-up. Middling results playing up at 4.0, bump-up.
Yet another guy on my team who was a mid-level 3.5 based on 2010 results. He also played up exclusively in 2011. Only played four of five matches, all doubles with a strong 4.0 partner. Lost all matches, some with competitive scores and some not. End result? Bumped up.
Bottom line: don't play at level if you want to be bumped-up; play up exclusively. And vice-versa: don't play up if you don't want to be bumped up; you might still be bumped up but you'd have to perform quite well for that to happen. A caveat: in the last 2-3 years USTA was making bulk "post-algorithm" adjustments to the year-end ratings. Specifics differed year-to-year as well as (it would seem) regionally, but overall the effect was of aggressively moving up players at the lower NTRP ranges. That was the case for 2.5, 3.0, and 3.5 in 2009, and for 2.5 and 3.0 in 2010 and 2011.