I have the exact same problem on my monitor. This is simply a different type of ghosting problem than what we're used to talking about.
My monitor at home is an Amptron G19FP. People keep hearing me say that it ghosts but the ghosting disappears magically when I play Starcraft. I didn't talk about what kind of ghosting it was before. But it's the exact same kind of ghosting that you have. When going from black to gray (any non-white color, really), it flips to white then slowly converges to the color I want. This shows up when I use office stuff because if I'm (say) scrolling on a webpage, the black previous text leaves a white image on the screen. Of course, playing Starcraft, there's not much black stuff so I never see this problem -- and I notice it only with black to another color, never gray to gray.
My monitor at home is not a TN panel. I don't know what panel it is, but I can say it's not a TN panel because the colors stay no matter the angle I look at it from...disappearing only when the bezel edge gets in the way. It's also too old to have overdrive, so that's not the cause either. I'm guessing that it's the same type of panel as the 2405FPW then. Is that suppose to be PVA/MVA or IPS (S-IPS)? I can never keep up with the Dell panels. Amptron's since taken it off their website, but you can Google it to see its stats and guess on the type of panel if you want (keep in mind that it's over a year old now, so the stats are naturally worse than now).
So what's happening is that the LCD overreact. We think of 0 -> 128 brightness as it just gradually going to 128, but it actually oscillates around 128, going over, under, etc., a couple of times, since the crystals physically rotate to change the brightness, and that rotation has inertia. Also, its inertia depends on the voltage state applied. In this case, it simply goes way over before simply settling back to the correct value; since my panel doesn't have overdrive, I'm guessing this is a fundamental property of the liquid crystal design itself. That means that this doesn't have to do with the response time per se (i.e. 16 ms or 12 ms), but the type of panel used. If you're wondering why this doesn't happen for white -> gray, it's because gray = no voltage, while white = yes voltage. LCD crystals move much faster going from a no-voltage state to a high-voltage state, but move slowly going from a high-voltage state to a no-voltage state. This means that when the liquid crystals are jolted to go to bright (from dark), it quickly reacts, and thus quickly goes over the ideal value; once over, it slowly settles down to the ideal value because it's slower going to a lower-voltage state. For white->gray it's already in the slower transition speed, so you don't have this problem (any overshoot is quickly taken care of by its quicker rise speed). Interestingly enough, if this is the case for this panel, that means these guys should come out with underdrive to take care of this =P
So yeah. After a while, I sort of got used to it and it never bothered me. It's a budget monitor though. Whether you want to do that with a non-budget monitor or get it returned is up to you.
As an aside, remember the old measurement method, black-to-white-to-black? As was noted on many websites, one's time is actually much, much faster than the other (something like 20 ms for black to white, but 5 ms for white to black, for a 25 ms response time). This is because for TN panels (which were the common type back then), black is actually full voltage, and white is no voltage -- hence dead pixels tend to be bright, not dark, and hence white->black is the fast one. For other types of panels, the opposite is true -- black is no voltage, and white is full voltage. So black->white is fast, while white->black is slow, for other types of panels (and dead pixels tend to be dark rather than bright). So why didn't this come up with TN panels? It's because they are inherently fast, despite color quality and viewing angle problems -- hence their early domination in the market. So while the effect happens with TN panels as well, people don't notice them as much. And this also means that reviewers should actually measure white->gray, not black->gray, for response times for the non-TN panels, as they are the slower type.
