No there is a correct solution regarding routing of de-coupliung capacitors and that is number 3, read up!
Please read up all there is on decoupling capacitors and how they work. It is inductance that is the problem, so creating the LOWEST inductance path makes the cap more effective.
Placing next to the ground pin does not make it any better for ground bounce (that is a myth), it is the total LOOP area that matters, the best placement is preferably equidistance between the pins you are de-coupling. It also depends on where the power planes are in the stack on whether it is better to place the capacitors on the same or opposite side of the board, whatever gives the LOWEST inductance.
To quote Henry Ott, "we are placing an L-C network between the power and ground, not a capacitor", thus it is a series resonant circuit, and as such once it passes any noise is above the circuits resonant point, inductance becomes the dominating factor and it doesn't work as a de-coupling capacitor. Looking at the systems requirements and engineering a solution is much better than random sprinkling of capacitors, and in a lot of cases (most!!) that is what happens, then you end up trying to treat the symtems (noise) and generaly end up adding to your woes.
For a bit of extra reading the problems and benefits of adding ferrite beads to the PDS equation.
You also need to choose a capacitor with the right parameters for decoupling, for the higher frequencies a low ESL is desirable, that means small package and inherant low parasitic ESL, also ESR is important, but not to low!
Howard Johnson - High Speed Digital Design, page 271 onwards.
Henry Ott - Electro Magnetic Compatability Engineering, page 431
Lee W Ritchey - Right the First Time, Chapter 34
Sorry if I seem to be getting a bit wound up, but decoupling capacitors are a bit of a hobby and years of having to work (even in this day and age) with EE's with their own myths regarding decoupling is a pain. Some engineers throw decouplers on a board like they were confetti, some use packages that are to big for the capacitor value and frequency's that are of interest etc etc. Start adding ferrite beads into the equation and your PDS (power delivery system) is goin to be noisy.