When I run it over me + all my LJfriends - all people in that set who only have one link into it (1848 users total), it finds:
two huge groups (not too interesting),
one medium-sized one:
ohyoshimi jproulx rahaeli absolutegorilla dannyvo theivyleague mesyeux kkchen jzzybelle crschmidt supersat hyounpark albertprime curree kierah jsu aieeeeeeee sarasu mea532 jaced xqzzy sherm zenithnadir rjpower odtgmdovh htang tesselation jcbarker rassilon42 thesuck ehsu harrchenone c_a_s_e evgnyg pixie_bebe rhennigade deveiant zanfur mechanyx mullenkamp ladydiamante zenspider tbgoldman elfbabe ratkrycek corivax asciident clinstabaw smirkingjustice blacklaw budindahizouse gothere unfurl drbrain lifeofbrian latinstorm xenofalcon bride thedutch sirnivek pezstar trelana jayo iseebi aquatwo sleepingcat karendipity burr86 manhattan flying_blind silenceleigh mrfurious jessical adrenalinegirl bertho voodootwo2 xb95 andrewducker toniotreschi cherrymoo jjhan khoi emmyhso justin2042 nyxie arie emmavescence joyofem sarah_mascara deslea yukatsuka _alita_ nineinchsilver vanillaice42 malerin evildave meggmoon kittenaskat starrygirlmehee alsatia eideteker roy braindrain thebubba aajwind antiques xtremesaints technodummy shahrizai spratt fuzzyfruit whit1206 virgaleo sashathumper cow metadaisy erinn thewildrose folk rzbryswrl tikva naruto_nerd tsukikoushi raventhon vanbeast neville janinedog dayar tenshisama rho acerbic mortaine fweebles ursamajor chrisg solcita jadine
that appears to be primarily LJ support and people from school, and then finally two small groups:
*** 313: lptran puzzlefite limenaide almondsilk rcyang supercat1102 ktnguyen qmdau
*** 328: diapholom hajime kayeyem xian nocte _generica jehilia soulwatcher bbt
The first one is my ex-roommate, his ex-girlfriend, and a bunch of his friends.
The second one includes two people I recognize from a group of friends in Australia. I don't know any of the others.
I think the problem is that by picking me and all of my one-hop friends, the nodes are unevenly distributed around me. There's also a low "q", the value the algorithm tries to maximize to indicate clustering strength. I wonder if throwing more data at it will make it worse or better...?
Anyway, if you'd like to play with it, the code (in C this time!) is up in my arch repository. I even put up simple instructions for checking things out for people who've never used arch.