They do a bit better at the top; the sample size at "top 10%" is getting small enough that tests are losing power, though:

R> lw <- read.csv("lw-survey/2012.csv")
R>
R> hpmor <- lw[as.character(lw$Referrals) == "Referred by Harry Potter and the Methods of Rationality",]
R> other <- lw[as.character(lw$Referrals) != "Referred by Harry Potter and the Methods of Rationality",]
R>
R> hpmor <- hpmor[order(hpmor$KarmaScore, decreasing=TRUE),][1:25,]
R> other <- other[order(other$KarmaScore, decreasing=TRUE),][1:83,]
R>
R> hpmortime <- hpmor$KarmaScore / as.numeric(as.character(hpmor$TimeinCommunity))
R> hpmortime <- hpmortime[!is.na(hpmortime) & !is.nan(hpmortime) & !is.infinite(hpmortime) ]
R> othertime <- other$KarmaScore / as.numeric(as.character(other$TimeinCommunity))
R> othertime <- othertime[!is.na(othertime) & !is.nan(othertime) & !is.infinite(othertime) ]
R>
R> sort(hpmortime, decreasing=TRUE)
[1] 506.78 300.00 283.96 203.62 138.46 133.95 117.61 115.92 72.22 66.67 59.09 50.00 36.92
[14] 35.05 35.00 33.90 28.60 26.67 24.82 23.96 20.36 19.28 17.71 11.91
R> sort(othertime, decreasing=TRUE)
[1] 1895.36 647.88 456.97 338.89 263.16 250.00 250.00 235.71 184.90 183.33 173.91 166.67
[13] 165.00 146.93 146.65 145.83 142.86 133.33 133.33 133.33 125.00 125.00 125.00 116.45
[25] 102.73 100.00 97.22 84.38 83.33 83.33 83.33 83.33 75.00 75.00 74.51 72.00
[37] 69.60 68.88 66.67 66.67 63.33 61.11 60.71 60.34 58.33 57.14 55.95 53.43
[49] 52.17 50.00 50.00 50.00 50.00 48.48 46.46 44.12 43.75 41.67 41.43 40.00
[61] 39.66 36.36 35.91 33.33 33.33 31.67 31.32 30.00 30.00 30.00 27.50 27.47
[73] 26.95 25.33 25.00 25.00 24.06 23.33 22.73 18.25 16.67 16.67
R>
R> t.test(hpmortime,othertime)
Welch Two Sample t-test
data: hpmortime and othertime
t = -0.544, df = 72.4, p-value = 0.5881
alternative hypothesis: true difference in means is not equal to 0
95 percent confidence interval:
-87.52 49.99
sample estimates:
mean of x mean of y
98.44 117.20