I like the way you think. I feel the same in many cases. If you want to see analysis on improvement of Colemak, check http://mkweb.bcgsc.ca/carpalx/
They give a lot of numbers. Keep in mind that your priorities on the layout arrangement might differ, so keep an open mind.

As far as I'm concerned Colemak is the way it is not only to be easy to transition from Qwerty. There have been discussions on how to optimize it and there isn't a single swap of keys that would make it better. There might be some more in-depth rearrangement that could improve it slightly, but I wouldn't bother, the difference would be very small if any and you lose the benefit of being similar to Qwerty and having convenient letter placement for shortcuts.

It's all about compromises. If you improve one aspect, you reduce something else. The biggest benefits of Colemak is the amazing home row usage and the very, very low same finger ratio (which are for me the two most important factors), as well as the low finger travel distance.

The way I look at it is that the easiness of transition from Qwerty is just a side benefit. It certainly wasn't the reason why I switched, but let me tell you – it helps.

- For one you learn it faster, even if you switch from another layout (I did from Dvorak), as long as you've touch typed Qwerty. Learning faster helps you reach your top speed and improve on it faster. So at any given point of time, you'd be typing faster than with layout that's "harder" to learn.

- As a second reason – it makes using a Qwerty computer easier. If you're daily Dvorak user, having to use Qwerty on others' computers is HARD. If you're Colemak, you can even almost touch type Qwerty with a couple of looks down at the keyboard. At least I can.

To make it harder but better, learn an AngleWide mod, useful lv3-4 mappings and an Extend layer (see my sig). Then learn Python or something... :)

The Angle mod fixes the J-B issue, and otherwise Colemak is actually so good that you should spend your surplus energy learning something useful.

I always did like that semi-colon placement. ... And the qu digraph now that I think about it. Never could justify moving minor keys though. Anyways, why not run it through the DDvorak site analyser? While I don't necessarily agree with its overall assessment of a layout, it is relatively easy to use, and gives some useful stats.

EDIT: code for DDvorak analyser, Capewell and the Modified Capewell layouts below. (Note: Do not paste Name into code section. Start with `123)

CAPEWELL
`1234567890-=
#.ymdfjpluq[]\
#aersgbtnio/*N
*Lxzcv;kwh,'*R
##*S#
*L
#######&**()_+
######JPLUQ{}|
######BTNIO?
######KWH<"
#
*R
~!@*#$%^
#>YMDF
#AERSG
#XZCV:
#

CAPEWELL EDIT
`1234567890-=
#.mydfjphuq[]\
#anesgbtrio/*N
*Lzxcv;kwl,'*R
##*S#
*L
#######&**()_+
######JPHUQ{}|
######BTRIO?
######KWL<"
#
*R
~!@*#$%^
#>MYDF
#ANESG
#ZXCV:
#

People have been doing that, but I don't remember the links.

I'll repeat and expand on what I said back when this topic was active: To make things harder but better, learn Extend (extremely nice!), WideAngle – and now you can throw in Curl which is actually a little harder because 1–2 more keys are moved but also better in some people's opinions (including mine).

I hope that Garquill eventually saw the advantages of Colemak, but if he didn't then I guess CarpalX is also a choice. Personally, I don't think his layouts are actually better than Colemak, but they're certainly harder! :-)