Xover frequency - am I deaf or what??

I all,
I have recently been playing with a variable active Xover instead of my fixed frequency (900 Hz) LCR filter in a 2 ways system. The top is a JBL LE85A in a 400Hz hypex round horn driven by a tube push pull, the low-mid/bottom is a LE15A in bass reflex driven by high current SS amp.
From info on this and other forums, and from speaker specs, I expected the LE85A to have to be Xover'd around 800Hz, yet, I can keep tuning the filter down to 550-600 Hz and I think that the sounds actually improve. Only when I get to around 360Hz does the sound becomes somewhat 'nasal' and there is a clearly aubible (to me) degradation.
What am I supposed to be listening to to find out if my Xover is set too low? I am starting to seriously question my listening skills!!.

The only 2 things that I consider as making this possible are that I do all my listening at 'socially acceptable level' (around 80-85 dB at the listening position) and that my horns are probably over specified (they are LeCLeach'h type and even for Fc of 400Hz they are bigger than a 300Hz tractrix when the full expansion is taken into consideration, the diameter is 46cm, about 18"), but this does not change the intrinsic properties of the LE85A...

Opinions?? Thanks much

JBL rates the LE85 for 500Hz operation, but at significantly reduced power. Due to it's high efficiency I wouldn't worry about that in a home stereo application. Horn cutoff frequencies are very confusing. Some rate them by the theoretical cut off and others by the lowest recommended frequency. I would want at least one octave between crossover and theoretical cut off.

Now about your variable crossover. I have recently discovered the gross inaccuracies of the screen printed numbers on the control panel and the actual crossover frequency. Since getting an accurate measurement system I checked the crossover frequency of my variable frequency active unit. It said 650Hz and it measured about 800Hz... I would expect most actives err on the high side to protect the drivers.

Widget

Thanks Widget, I wondered as well about the accuracy of the markings as the scale is clearly not linear yet the separations between numbers are, it is not possible to guess the 'true' values of the intermediate steps.