There's a pretty good description of this on Wikipedia's site. Just search for square wave. Here's an animated gif from that site:
As it turns out, the fundamental Fourier component has a peak level that is nearly 2.1 dB above the peak level of the square wave. The exact ratio is 4/
π.
I should also point out that a square wave is the worst case waveform for these types of effects and that you'd only see a 2.1 dB increase in peak level if you used a brick wall filter. Of course, no such thing exists, so that's not a big concern. However, if you use a low pass filter with high Q poles, you can get a lot of ringing at the rising and falling edges of the filtered square wave and that can cause a huge increase in peak level, especially if the fundamental frequency is harmonically related to the cuttoff frequency of your low pass filter. On the other hand, that's not the type of filter you'd be using for equalization of audio sources. More likely you'd be using a parametric equalizer with relatively low Q stages, so ringing would not be a concern.
So yes, it's definitely possible to increase peak levels while actually reducing the rms power in the signal by reducing the level of a certain portion of the signals spectrum, but it's not at all likely to happen in the real world when equalizing raw, uncompressed, unclipped, unlimited live recordings like we do here.