Energy flow:

Energy fractions in a tipical jet:

Charged particles (including e,μ): 60%
Photons (mostly from π0, also brehmstralung): 25%
Neutral hadrons (KL, neutrons): 10%
Neutral V0's: 5%

The long-lived neutral hadrons are the only component that you cannot improve wrt an HCAL-only measurement.
The jet angular resolution is also limited by these neutrals (since for charged tracks and photons the angular resolution is much better than HCAL).

If you use only the charged tracks to form jets (and extrapolate to the rest) you have a jet energy resolution of 25%-30% (independent of energy): so use of calorimeter is nevertheless essential.

Ideal "particle flow" (i.e. ideal case in which all particles can be reconstructed and identified):
- charged particle + V0's (=65%) have infinitely good resolution
- photons (25%) have the ECAL resolution (18%/√E in ALEPH)
- neutrals (10%) have the HCAL resolution (90%/√E in ALEPH)
- in this ideal case an e/π ratio can be applied (compensation)
You obtain in the ALEPH case an energy resolution of 3.5%, despite the modest calorimeter resolution.
In the real world, the ALEPH energy resolution with energy flow was 7%. The difference comes mostly from the poor neutral identification efficiency and purity, which comes from the coarse HCAL granularity.

Requirements:

Efficient tracking, even without pixels (since you need the V0's!)
Large magnetic field (to separate charged and neutral particles in the calorimeter)
High ECAL granularity (to separate photons from π0's, or π0's from π+'s in tau decays, or brems from electrons)
Efficient and pure electron and muon id, down to small momenta
Good HCAL granularity (not true in ALEPH, nor in CMS...)

The critical parameter is not the energy resolution, but the angular one.

ILC design takes that into account, with extremely fine granularity for ECAL and HCAL.

CMS plus and minus wrt ALEPH:

(+) larger magnetic field (x3)
(+) magnet is outside HCAL (very little energy loss between ECAL and HCAL, potentially better link between ECAL and HCAL clusters)
(+) preshower
(=) much better ECAL resolution (but it is not critical)
(=) worse HCAL resolution (but it's not critical; what matters is the neutral hadron identification capability)
(-) coarser HCAL granularity
(-) non-linear response to hadrons (charged hadron energy linear subtraction might be less efficient)
(-) no longitudinal segmentation in ECAL (worse electron and photon purity)
(-) no redundance in energy measurement from ECAL and HCAL (no possibility of noise cleaning)
(-) more material in front of the calorimeter: more photon conversions (good if e+e- tracks can be reconstructed), more nuclear interactions (some energy is lost, neutrals and charged hadrons are scattered at large angles; can be recovered in part if secondary vertices can be reconstructed)
(-) larger jet momenta (worse separation between hadrons): with increasing Pt, energy flow tends to purely calorimetric
(-) pileup