Crystal oscillators are used because the stability quoted is the overall stability that you get. You don't have to worry about what load capacitors to add or the tolerance or those. You only need one pin on the PIC for the clock and layout is not important.
With a crystal, the exact value of the capacitors is very difficult to predict, because the time delay though the oscillator inverter will alter the frequency, so you need to adjust the capacitors to take account of that. It is almost always done with some trial and error on the first production boards. The oscillator inverter on an IC can often have too much gain, and some crystals will oscillate at overtone when they are not supposed to. That can be a real problem with low frequency crystals, such as 4 MHz ones. The crystal inside will be quite thick compared to its length, and it has to be cut down to a thin strip to get in the case. Both of those cause problems such as difficulty in getting the oscillators to start, or running at spurious or overtone frequencies. Also those crystals are adjusted to +/- 50 ppm, and they have a +/- 50 ppm stability. You will only get +/- 50 ppm over the temperature range if you have some way of adjusting each one to the right frequency at room temperature.
Where I used to work, we made oscillators and crystals. A lot of customers had a lot of trouble getting crystals to run at the right frequency, and it is simply not worth the effort unless production runs are large, or there is some odd requirement.
On one application at 4 MHz, I used a crystal in an HC33 package, which is 19 x 19 x 8 mm, so that I could get one with a lower equivalent series resistance, and more predicable temperature characteristics. 4 MHz in the surface package will be a far more difficult crystal to work with.