You might consider an array of transducers, if you drive them all in phase you will end up with a narrow beam (effectively a phased-array sonar). The wavelength of a 40kHz sound wave is 8.6mm, so placing those MaxSonar transducers as close together as you can (about 16.4mm) gives an array spacing of approximately 2*lambda.

The first zero in the transducer array pattern is at (sin theta = lambda / ((2*N +1) * d))) where d is the distance between the array elements and N is the number of elements (linearly). So for d = 2*lambda (ie. the closest you can space these transducers):

For N=2, theta = 5.75 degrees, or beamwidth ~11 degrees.

for N=4, theta = 3.2 degrees, beamwidth = ~6 degrees.

for N=8, theta = 1.7, beamwidth = ~3 degrees.

Note that these are array factor improvements, the MaxSonar transducer looks like it has a beamwidth of around 30 degrees, this value is modulated onto the above improvement factor. The take-home result is that doubling the number of array elements results in an approximate halving of the effective beamwidth.

If you were trying to steer the beam electronically, you would get severe grating lobes due to the wide array spacing. If you're only going to drive the whole thing in phase (ie beam on the boresight, you don't need to worry.