Like many here, I set up a bus, add a 100% wet reverb plugin to that bus, and set up a send on the armed track to that bus.  All other instruments are bussed to instrument groups, and all buses have sends directly to the headphone out on my interface (Focusrite Saffire Pro 40).  I leave Input Echo on, as my buffer settings are typically 128 when tracking and latency isn't an issue.  The only catch is that if I'm layering multiple vocal tracks, I turn down the reverb send on all vocal tracks except for the one I'm recording, otherwise it'll get pretty mushy with all of those vocals hitting the reverb, which I typically set pretty aggressively for tracking.
 
There are probably easier ways to do it, but after experimentation this method works well for me.

Here's my no-hardware trick for this, that works even in high-latency scenarios:
 
1/ Direct monitor the vocalist through your audio interface
2/ ALSO enable Input Echo on the track you're recording to, but turn the volume to -inf in Sonar
3/ Add a send to a reverb bus, and make post-fader
4/ Make sure the reverb on the bus is set to 0% dry, 100% wet
 
Voilà, direct dry signal, with no latency, but also reverb in the headphones, and no reverb printed. There will be latency on the reverb itself, but in most cases it won't matter.

I have an RME UFX so use its onboard DSP for any monitor fx for vocalists or other.

RMEs Totalmix Mixer and Matrix, allow any routing you can imagine, completely independent of Sonar.

You can setup 8 completely different recallable mixes all with their own ew, reverb and dynamics settings. The closer I can make the vocal sound to what it will do in the final mix, the better performance I get from the vocalist. This makes getting the monitor mix right, a very important step and saves lots of work in the final mix

Sacalait
I don't have a mixing board.  I'm using Sonar only.  This is what I've figured out and it works on reverb- pretty much regardless of what settings you have in the mixing latency (well, at least 1024 or less).

1:  In the master section I insert a stereo bus and drag in a reverb plugin (i LIKE the Lexicon plugin that came with 8.5 and X1!).  Name it "Reverb"
2:  Insert an audio track and insert the Reverb send
3:  On this audio track click input echo to "On", lower the volume all the way to OFF (vital because it eliminates the dry signal), assign the input to the mic you're using on the vocalist, click the reverb to "pre fader", and adjust the send level to taste.

Because this is a reverb effect you wont notice latency.  In fact if you have the audio buffer setting to higher (like 1024) it almost helps because of the pre-delay.  NONE of my artists EVER complain about latency when I do it like this.  I don't think this would work with effects like EQ, compression, or really anything else but reverb but it works nicely to give an ITB reverb mix to the performer.

Anyone else do it this way? 
 

This is how I did it before I got my RME with DSP reverb. Had no problems with it.
All the best.

This is why I moved from the RME Fireface (and outboard) to the Apollo. Being able to put something close to my vocal chain for low latency monitoring and then add some FX for the artist is a godsend ( I don't print any of the procs/FX ) . I think there are more options for DSP processing for monitoring via interface now-a-days.

Yeah the newer RMEs have it too, or at least my UCX does. So far I've always been able to rely on decent interface mixing apps. My E-MU 1820M also had a basic reverb and plenty of routing options.

I've had trouble in the past whenever I tried to "bypass" direct monitoring and go for software monitoring, vocalists can get really particular about quite low latency.