OK, here's the frequencies for the last 200 BTC blocks.... as I suspected, the outer ends are more heavily weighted:

---

Frequencies	last 200	percent
10	25	12.5
20	17	8.5
30	20	10
40	20	10
50	19	9.5
60	19	9.5
70	14	7
80	17	8.5
90	24	12
100	25	12.5

---

Here's the same data broken down by quartiles:

Frequencies	last 200	percent
25	55	27.5
50	46	23
75	43	21.5
100	56	28

---

Admittedly, that's only the last 200 samples, but the analysis is disturbing... this represents (at 4 blocks per day) about 50 days of data.
BTW, to be clear, the table values mean the actual number of data points between the bin boundaries, and their percentages. For example:

25

55

27.5

means that, in the range of 0 to 25%, there were 55 samples; and that represents 27.5% of the 200 samples in the dataset...

I'd have to be more of a statistician (or an actuary) to know how readily the statistics of small samples would allow this kind of variance, but perhaps it's something someone should look into, in case there's an as-yet-undiscovered peculiarity in the algorithms?

-- DickMS