If a plug-in has latency, it's there for a reason. Often times it's lookahead, which increases the precision of dynamics control. Other times it's just because it uses a lot of CPU. Transient Shapers often require both. If you compare a low-latency TS withone that has a lot of latency, odds are you'll favor using the one with more latency.
BTW it's
not a given that linear-phase always sounds better. At high Q settings, there can be pre-ringing. This is one reason LPs are ideal for mastering, where you want "surgical" EQ and don't need to push the Q much. LP makes the biggest difference at high frequencies but also, remember that the band interactions and phase shifts with conventional EQ are what give them a specific "character." People pay for EQs that emulate vintage console EQs specifically because they DO have anomalies. Also, all things being equal, linear phase EQs will seem to have less of an effect compared to similar settings in a non-LP EQ.
I think a lot of the reason people hear big differences among EQs is because there
are differences, due to design philosophy and intended goals. For a quick reality check, put the QuadCurve on program material and try out the different curve types. Imagine if each one was a separate EQ module instead of options in a single EQ module. Odds are people would debate which EQ sounds "better."
