Practice the speed of your algorithms specifically.
Learning faster algorithms is great, but when people are practicing speedcubing, the single most important thing people overlook is that the majority of their time is spent executing the algorithms.
Increasing the speed with which you execute your algorithms will drastically increase the speed at which you solve the puzzle, regardless of the algorithms you're using. The beginner's method takes me about 30-40 seconds to execute - and I'm not using OLL and PLL. I did it simply by increasing my speed of execution. Then I learned OLL/PLL.
Take the algorithms and steps you know, and push the speed faster. Time just the execution of one algorithm (or, if it's too fast to time, instead time several iterations of it). Work on increasing the execution speed of the algorithms you're using. Determine which ones take the longest, and practice those.
For instance, here's an example using a couple of the slow algorithms I created (though you might want to do something like ten tests):
Edge 3-cycle | 3 iters | 9.27 | 10.86 | 10.54 | 9.99 | 10.40 |
Corner 3-cycle | 3 iters | 12.76 | 10.79 | 11.64 | 10.64 | 10.34 |
Therefore, the edge 3-cycle takes me an average of 3.4 seconds, and the corner 3-cycle takes me an average of 3.7 seconds. (These times are very slow, but that's primarily because I'm seriously out of practice and using a poorly-lubricated, old cube.)
- Index the steps and algorithms you know (for instance, make the cross a step and F2L a step, then time specific last layer algorithms)
Identify sequences that seem abnormally slow, or abnormally inconsistent.
You can do this with a standard error. From the above, my edge-cycle error is 0.244, and my corner-cycle error is 0.392. This betrays the fact that corner-cycles are drastically less consistent than edge-cycles.
A second run of 5 shows that I'm only a little out of practice, and that I really should relubricate and retension this cube. My average time for the corner-cycle dropped 0.5 seconds, and the error dropped to 0.273.
Practice those sequences' execution times until they're much more consistent.
- Try to identify where you might be getting caught - is one of your rotations going too far? (Chances are, at least one rotation is consistently out of alignment.)
This method should help you identify the slow points of your solution and improve those specifically. Then, practice tying the sped-up methods together.
Your visual processing time can't easily be trained, and that's just something that's going to drop the more you solve the cube.