Why is the difference between the Atitool and Furmark so large (124 - 87 = 37)?
For other cards it's about 10%, for this one over 40%! I don't question your results, but I find it quite strange.
Cooling 87W is easy with AC S1 and undervolted Turbo module, but 124W is difficult IMO (or noisy - not an SPCR choice).
PS. How about retesting with Furmark for other high power cards, like HD4850 and GTX260?

Seems like a predictable outcome. Sparkle is the cheapest brand on newegg, and their card is the crappier of the two varieties. It's the longer card, needs two power connectors, and has a loud heatsink. My Gigabyte 9800GTX+ with factory Zalman cooling is loud like a tornado at full speed, but peaks at something like 54 degrees. Undervolting the fan to 5v makes it inaudible over the PC's regular noise and brings the peak ATITool temp to 74C - roughly the same as the Sparkle's temperature while churning out tornado loudness. It seems like another case of you get what you pay for.

EDIT: OK, the GTS250 is officially one of the craziest re-brands yet. Originally, the GTS250 was supposed to be at least in line with the nvidia reference specs. The key features being 9.0" long, single 6-pin connector. I just went through NewEgg's pictures and using a simple ruler determined there's no fewer than 3 versions being sold as a "GTS 250":

BFGTech GTS250 = 9.0"
XFX GTS250 = 9.0"
Sparkle GTS250 = 9.5"
evga GTS250 = 9.5"
zotac GTS 250 = 10.5"

I have a feeling only the 9.0" ones are genuinely new cards, the others are just re-labeled 9800GTX+. The evga GTX+ for example was always a 9.5" card with a single 6-pin power connector, so for evga they could simply slap a new sticker on the GTX+, flash it with a new name in BIOS and they're done, so long as nobody cares about the extra half inch. The Zotac is the funniest, since they make their card the least convenient of all, requiring a single 8-pin connector. It's like they get to just make things up as they go along, lol!

Indeed, some of the gts250 including gigabyte`s are reusing 9800tgx pcbs. That`s not necessarily a bad thing, most of those custom cards consume about 10 watts less than the reference 9800 design which brings them in line with the gts250 reference board.

Yes, sparkle is one of the cheaper makers, but overall this card seems decent: modest power consumpion and a good price to offset the need for aftermarket cooling. I would still prefer something like the gigabyte gts250 though since it comes with an improved zalman cooler.

The Gigabyte is the shortest card too; even my 9800GTX+ is only 8.3" in length. It's still not perfect though; it doesn't exhaust heat out the back of the case, nor do they use full-speed memory on the 1gb model. At least for the GTX+ it's only 2000mhz, as opposed to 2200mhz for the reference design and most implementations. I'm not sure if Gigabyte increased it to 2200mhz or not for the GTS 250, but I somehow doubt it considering they fail to mention the memory speed anywhere in the product specifications on their website. Not a real big deal, but it still means it is slower than the others by a little bit.

I've read that Sparkle provides an under/over-clocking utility called SPA Tune OC, it'd be nice to see how much IDLE power consumption is there when the card is set to Green mode (lowest frequencies).
Normally i'd set that mode while in 2D mode (working in Windows without any "advances" application) and only set the default frequencies when playing a game,etc (or maybe not, depends on how much the 3D is complex).

Also note that almost every manufacturer overclocks their cards at the factory these days.