This answer is based in part on my personal experiences in building solar powered LED lighting in China over the last 4 years. You learn a few things along the way :-).
The actual lifespan, properly operated, depends critically on the manufacturer.
The "big 5" (or 6) and their licensees can make long life LEDs. Vanishingly few others manage.
If it's Luxeon / Lumileds (Philips), Osram, HP (Avago) , Nichia, Cree and probably Siemens, Sharp, Seoul semicondutor* OR a company that has licenced their technology and used it properly and completely then you may well obtain the performance claimed. (I will have missed a few here).
If there is NO traceability by formal contract with the big 5 you will very very very likely not achieve the claimed performance.
Up to 10% over rating may not make a vast difference but it will definitely be a measurable difference. eg 800 for your 750 mA part is about 6.5% over. 850 mA is 13%+. The differences between the two will be noticeable.
LEDs of another age could often withstand pulsing at immense current ratios as long as mean power was limited. Modern "phosphor LEDs" often have I_abs_max very little more than I_optg_max.
Die temperature and absolute current level both impact lifetime wholly independently. (Many people are unaware of this an seem to think that temperature is what matters and that current is only a "proxy" for temperature.
As a guide, a part I use that is rated at 50,000 hour at 30 mA is rated at 14,000 hours at 50 mA.
Here's a LED patents relationship chart.
ex LEDs magazine 2006/2007 so somewhat dated.
Note that using a manufacturer's die may NOT mean you get all the advantages. eg many makers say "Cree chip inside" BUT they can do things that reduce lifetime. Lead frames, epoxies, mounting stresses, ... all have effects.
This Future lighting solutions page may be useful. Luxeon features in the fine print
Lumileds do a very nice lifetime versus various parameters document with good discussion. I can dig a coy up in due course if you don't find it by web or site search first.