Minimising the length of I/O lines will keep parasitic capacitance and inductance low.
This is the best overall general advice you will get regarding I/O lines.
However, I/O lines on PICs have characteristics that depend on whether they are programmed as
digital or as
analog lines. Additional capacitance on digital I/O is almost always undesirable because it slows the rise time and fall time of digital edge transitions. This in turn leads to other undesirable behavior, such as timing errors and even missed transition-events (bandwidth limitation). It is best to avoid adding any additional capacitance to digital I/O lines by carefully following Steve's recommendation. Circuit board layout is important here.
PIC I/O programmed as analog inputs are sometimes benefited by an RC circuit attached near the appropriate pin, the time constant being chosen to roll off the high-frequency response of the input to minimize analog signal noise and prevent "false positives" in comparator applications. An RC network (series input resistor plus a shunt capacitor) will attenuate fast-changing analog-to-digital converter inputs, which can lead to inaccurate conversions for rapidly changing analog inputs, but with decreasing digitization noise. Such "signal conditioning" should be performed in earlier stages, where it can be better controlled, with the A-to-D input driven from a low-impedance, low-noise, voltage source. Placing an RC network at the input to the A-to-D is a "band aid" that sometimes works.
Long circuit paths are invariably associated with increased parasitic capacitance and inductance. While a few additional picofarads, or a few extra nanohenries, here and there may not seem like much, they are not part of an intentional design. It is good engineering practice to minimize such parasitic capacitances and inductances.
Back in the day, some people would twist short lengths of insulated wire together to make a "gimmick" component and then prune the gimmick to such a length that it served its intended purpose. While this generally "worked" for one-off projects, it was not good engineering practice because it lacked repeatability.
