Yeah, carbon comp is notorious. Film is fine, I've got tons of them and have never measured one in error -- even a few I've toasted in the course of breadboarding....
Carbon composition has the best pulse-overload/surge power handling, so they are still used and for sale.
Their parasitics- inductance and capacitance are quite low as a slug. I took apart a Philips CR37 220k 1/2W carbon film and it was only 8 turns spiral and rated past 10MHz before reactance causes error beyond its tolerance. I think metal film resistors have higher turns (inductance).
And note that the
capacitance between those turns is what's dominating. Resistors over 200Ω or so tend towards capacitive reactance at high frequencies, while resistors below 50Ω or so tend inductive. Simply the ratio of DC resistance to characteristic impedance of the component. It's hard to make an axial (flat to the board, not standing) or chip resistor with an impedance much different from this.
The spiral cutting will tend to increase characteristic impedance, at least over the modest range where the spiral's length is dominant (low GHz?).
There are wide-style chip resistors, which can have lower characteristic impedance. Good choice for current shunts. Or just use a bunch in parallel, connecting with wide pours.
Wirewound resistors, somewhat obviously, have a lower bandwidth, but which way that swings isn't actually all that obvious.
The basic material property in play here is the ratio of resistivity to inductivity of the material: the bulk time constant, as it were. A longer time constant (lower resistivity) means inductance takes over at a lower frequency. This is a good reason to avoid using PCB traces as resistors -- besides the horrendous tempco and poor fab tolerance, they go inductive above perhaps 10s to 100s of kHz.
A high resistance material, or bulk equivalent (carbon granules are pretty conductive, but carbon composition has poor contacts and random paths between grains, so approximates a high resistivity material in bulk), is attractive for resistors, for this reason.
Films can also be made ever thinner; a PCB with 35um copper isn't very good, but if you could get one with say 1um, it would be passable (assuming the thin layer doesn't corrode away, of course).
That's why metal film (thick and thin) are pretty well standard.
Anyway, wirewound -- because the metal has a longer time constant, and on top of that, the winding design tends to emphasize inductance (i.e. a helix). There are more turns than a film resistor needs, so the characteristic impedance is high and the cutoff frequency is low (typically low MHz). Large values (>1k, 10k?) will again tend to be capacitive, while small values tend to be inductive. Noninductive (Ayerton-Perry) winding helps cancel most of this. (Note that the inductance due to sheer component length is unavoidable.)
Tim