Your premise is wrong. Linear phase filters offer linear phase.
Phase is not the same as time delay – at $f_1$, time delay $t$ will lead to a phase of $\varphi_1=-\omega_1\cdot t$, whereas at $f_2$, the phase will be $\varphi_2=-\omega_2\cdot t$ ($\omega$ is $2\pi f$); as you can immediately see, phase is a linear function of frequency.
(It's always good to remind oneself that frequency is just the derivative of phase over time, $\omega=\frac{d\varphi}{dt}$, and thus phase is just an integral of frequency over time.)