If you use Oracle Solaris 11.3, you can have persistent L2ARC (survives reboots). illumos and FreeBSD do not yet have this feature, but it is on the roadmap.
Aside from that I would not look at boot times too much. The difference between a slow HDD boot and a fast SSD boot could be 30 seconds to 3 minutes. If you set your workstation (BIOS setting) to automatically boot 5 minutes before your day starts, you will see no difference. Even if you wait, it is at most once a day.
You did not write about your workload: how large is it, how much random IO is there? Caching is a combination of most recently used (MRU) and most frequently used (MFU). For example, compiling a big codebase or working with a large training data set or using a database will quickly populate the cache either way.
Also, be advised that your first solution has different needs on the same disk: SLOG devices need very low latency, very high write endurance, acceptable random write speeds and do not need to be bigger than 4-8 GB. L2ARC devices on the other hand need to be much much bigger (depending on your RAM), need fast random writes and not much else. If you want to combine this, you will find yourself in very expensive territory (essentially, best of both worlds). It may be more efficient to split and use two different devices.