The non-spoiler plot summary: Police are called to investigate a naked man with a shotgun in a New York City apartment. Things go very badly. As part of this, the police break into an apartment and find it's full of guns. Hundreds, maybe thousands of guns. Each one seems to have been used in an unsolved homicide, one apiece. They're not just lined up, they're carefully arranged in some clearly incomplete pattern.
Some of the guns are over a century old.
Ellis handles all of this cleverly, sensibly, and coherently. There are apparent red herrings everywhere, yet each turns out to be either literally true or completely fair. Point builds on point logically and compellingly. The resolution is satisfying and self-consistent - something I didn't think he (or anyone else) would pull off given the situation lain out.
The pacing is brilliant. After a bit of scene setting, the action slowly ramps up to a huge climax that several times had me going "no, no, he's not going to do that" followed by him doing exactly that - but not always in the way expected, or for the reason expected.
Ellis fills this book with a fascinating and varied cast of characters. Even the minor ones are fully realized. One doesn't expect a lot of character development in the course of a single relatively short (310 pages) police procedural, but some of the major ones progress in interesting and believable way. The dialog is excellent, and with the exceptions of a few Britishisms, it all rings true.
This isn't to say there aren't problems. Ellis' descriptive prose is weak and sometimes clumsy, possibly because he's mostly been writing comic books (really really good ones) and has been able to rely on the artist to provide that sort of description. But given the improvement from Vein to Gun Machine, I suspect we'll see further improvement in his exposition as well. Hell, for all I know that exposition got better as the book went along and I was just too wrapped up in the people and plot to notice it.
How much did I like it overall? I wish his next book would be published tomorrow. Strongly recommended.