The other two things are totally valid. Tcl has no references, so you can't build your own linked list. And all values are (at least semantically) strings, so numeric operations are slow.
But of course, a lot can be done to optimize those things in the implementation, and over time at least some of that optimization was done.
All values are _still_ semantically strings (it's a core tenet of Tcl: "Everything is a string (EIAS)"), and in 1994, they were practically strings, too (that was the implementation). In 1999 w/ Tcl 8.0, Tcl became byte-compiled and adopted "Tcl_Obj" to represent values[0]. The Tcl_Obj (nothing to do w/ object-oriented programming), is what's called a "dual ported" object. It contains a string-representation of it's value (to preserve EIAS), and a "native" value of it's last computed use. For example, if you:
% set a 9.0
% expr { $a * 0.01 }
The native value of the Tcl_Obj $a refers to will be a float. Repeated calls to $a in a float-context will continue to use this already-computed native value. No need for further string interpolation.[0] http://www.tcl.tk/software/tcltk/8.0.tml
[edit: formatting]
I've had cases where opening a 'C' program as a pipe and doing heavy number crunching in the 'C' program outperformed native Tcl.
That all seems normal to us now, but it was uncommon back then.
Was that the case in 1994, or are you writing about the current state of Tcl?
I didn't get to it until around v7.x around 1995 and lists feature prominently in Osterhout's book.
Did I miss the rebuttal to this?