For my money, here are some of the more evident flaws with 3.5:
*Excessive complexity in generating creature and character stats (too many sources of bonuses, too many skill points, etc); John Cooper's unofficial errata in his reviews on this website are proof enough of this - if the core designers and developers for the system have trouble generating rules-legal stat blocks, something has gone seriously wrong;
*Some overly complex resolution systems (grapple, aspects of AoO, rules for non-lethal damage, unarmed attacks, overlapping mechanics of spell resistance and saving throws, contrast between full-action spells and one round spells, to name a handful);
*A degree of incoherence in the relationship between the metagame and in-game aspects of rewards: treasure is both a reward for players (as it improves their PCs) and also has a clearly defined in-game significance, but XP are far more ambiguous - at times the suggestion is that XP, levels etc are purely metagame concepts (and this is also suggested by the fact that they are earned in a way that can only be given a metagame justification, namely, by adventuring) but the spell component and magic item rules treat them as an in-game character resource also;
*Coherence problems also in the ways characters/creatures are modelled - hit points, BAB etc suggest a focus on high fantasy, but skill definitions and rules for skill use are much closer to RM or RQ-style simulation of gritty fantasy;
*Related to the above, a lack of rules to facilitate fairly common tropes of high fantasy, such as pursuits, acts of derring-do, and characters pushing the limits of their power (either physically or magically) and exhausting themselves as a result - in part this is a consequence of the absence of metagame mechanics for giving players a degree of narrative control over the outcome of PC actions;
*The use of Raise Dead (which has in-game as well as metagame significance, and is subject to sometime arbitrary GM interference) rather than Fate Points (a purely metagame device) to give players narrative control over the fate of their PCs;
*An inability to decide whether alignment is purely descriptive (as the PHB suggests) or also prescriptive (which is implied by the assumption in most modules and campaign worlds that PCs are Good), and a related inability to coherently explain the internal psychology of "ordinary" Evil people (as opposed to serial killers and other psychopaths).
Obviously none of these make 3.5 unplayable, given the amount of play that it sees. But indications are that 4E will tackle many of these issues, and that will make it a better set of rules.
A top-quality set of rules doesn't need parts that get ignored or applied in a haphazard way. The rules should play smoothly as written. This sort of clunkiness was typical of AD&D, but one aim of 3E was to get rid of it. Where 3E failed, 4E might have a chance of success.